Download Intro - Stanford Secure Computer Systems Group

CS140 – Operating Systems
Instructor: David Mazières
CAs: Varun Arora, Chia-Hui Tai, Megan Wachs
Stanford University
– p. 1/2
Administrivia
• Class web page: http://s140.ss.stanford.edu/
- All assignments, handouts, lecture notes on-line
• Textbook: Operating System Concepts, 7th Edition,
by Silberschatz, Galvin, and Gagne
• Staff mailing list: [email protected]
- Please always email staff mailing list for help
• Newsgroup: su.class.cs140 ← main discussion forum
• Key dates:
- Lectures: TTh 4:15-5:30, Gates B01
- Section: Some Fridays 4:15-5:05, Gates B01
- Midterm: Thursday, October 25, 4:15-5:30pm
- Final: Wednesday, December 12, 12:15pm
– p. 2/2
Lecture videos
• Lectures will be televised for SCPD students
- Can also watch if you miss a lecture, or to review
- But resist temptation to miss a bunch of lectures and watch
them all at once
• SCPD students welcome to attend lecture in person
- 4:15pm lecture time conveniently at end of day
- Many parking spaces don’t require permit after 4pm
• Other notes for SCPD students:
- Please attend exams in person if possible
- Feel free to use newsgroup to find project partners
– p. 3/2
Course topics
• Threads & Processes
• Concurrency & Synchronization
• Scheduling
• Virtual Memory
• I/O
• Disks, File systems, Network file systems
• Protection & Security
• Non-traditional operating systems
– p. 4/2
Course goals
• Introduce you to operating system concepts
- Hard to use a computer without interacting with OS
- Understanding the OS makes you a more effective programmer
• Cover important systems concepts in general
- Caching, concurrency, memory management, I/O, protection
• Teach you to deal with larger software systems
- Programming assignments much larger than many courses
- Warning: Many people will consider course very hard
- In past, majority of people report ≥15 hours/week
• Prepare you to take graduate OS classes (CS240,
240[a-z])
– p. 5/2
Programming Assignments
• Implement parts of Pintos operating system
- Built for x86 hardware, you will use hardware emulator
• Four implementation projects:
- Threads
- Multiprogramming
- Virtual memory
- File system
• First project distributed at end of this week
• Attend section this Friday for project 1 overview
• Implement projects in groups of up to 3 people
- Pick your partners today
- Lecture will end early so that you can do this
– p. 6/2
Grading
• 50% of final grade based on midterm and final exams
- max (final, (midterm + final) /2)
• 50% of final grade based on projects
- For each project, 50% of grade based on test cases
Please, please, please turn in working code, or no credit
- Remaining 50% based on design, outlined in document
• Do not look at other people’s solutions to projects
• Can read but don’t copy other OSes (Linux,
Open/FreeBSD, etc.)
• Cite any code that inspired your code
– p. 7/2
What is an operating system?
• Layer between applications and hardware
• Makes hardware useful to the programmer
• [Usually] Provides abstractions for applications
- Manages and hides details of hardware
- Accesses hardware through low/level interfaces unavailable to
applications
• [Often] Provides protection
- Prevents one process/user from clobbering another
– p. 8/2
Why study operating systems?
• Operating systems are a maturing field
- Most people use a handful of mature OSes
- Hard to get people to switch operating systems
- Hard to have impact with a new OS
• High-performance servers are an OS issue
- Face many of the same issues as OSes
• Resource consumption is an OS issue
- Battery life, radio spectrum, etc.
• Security is an OS issue
- Hard to achieve security without a solid foundation
• New “smart” devices need new OSes
– p. 9/2
Primitive Operating Systems
• Just a library of standard services [no protection]
- Standard device drivers, interrupt handlers, I/O
• Simplifying assumptions
- System runs one program at a time
- No bad users or programs (often bad assumption)
• Problem: Poor utilization
- . . . of hardware (e.g., CPU idle while waiting for disk)
- . . . of human user (must wait for each program to finish)
– p. 10/2
Multitasking
• Idea: Run more than one process at once
- When one process blocks (waiting for disk, network, user
input, etc.) run another process
• Problem: What can ill-behaved process do?
– p. 11/2
Multitasking
• Idea: Run more than one process at once
- When one process blocks (waiting for disk, network, user
input, etc.) run another process
• Problem: What can ill-behaved process do?
- Go into infinite loop and never relinquish CPU
- Scribble over other processes’ memory to make them fail
• OS provides mechanisms to address these problems
- Preemption – take CPU away from looping process
- Memory protection – protect process’s memory from one another
– p. 11/2
Multi-user OSes
• Many OSes use protection to serve distrustful users
• Idea: With N users, system not N times slower
- Users’ demands for CPU, memory, etc. are bursty
- Win by giving resources to users who actually need them
• What can go wrong?
– p. 12/2
Multi-user OSes
• Many OSes use protection to serve distrustful users
• Idea: With N users, system not N times slower
- Users’ demands for CPU, memory, etc. are bursty
- Win by giving resources to users who actually need them
• What can go wrong?
- Users are gluttons, use too much CPU, etc. (need policies)
- Total memory usage greater than in machine (must virtualize)
- Super-linear slowdown with increasing demand (thrashing)
– p. 12/2
Protection
• Mechanisms that isolate bad programs and people
• Pre-emption:
- Give application a resource, take it away if needed elsewhere
• Interposition:
- OS between application and “stuff”
- track all pieces that application allowed to use (e.g., in table)
- on every access, look in table to check that access legal
• Privileged/unprivileged mode:
- Applications unprivileged (user/unprivileged mode)
- OS privileged (privileged/supervisor mode)
- Protection operations can only be done in privileged mode
– p. 13/2
Typical OS structure
P1
P2
P3
P4
user
kernel
VM
file
IPC system
sockets scheduler
TCP/IP device device
device
driver driver
network
console
driver
disk
• Most software runs as user-level processes (P[1-4])
• OS kernel runs in privileged mode [shaded]
- Creates/deletes processes
- Provides access to hardware
– p. 14/2
System calls
• Applications can invoke kernel through system calls
- Special instruction transfers control to kernel
- . . . which dispatches to one of few hundred syscall handlers
– p. 15/2
System calls (continued)
• Goal: Do things app. can’t do in unprivileged mode
- Like a library call, but into more privileged kernel code
• Kernel supplies well-defined system call interface
- Applications set up syscall arguments and trap to kernel
- Kernel performs operation and returns result
• Higher-level functions built on syscall interface
- printf, sanf, gets, etc. all user-level code
• Example: POSIX/UNIX interface
- open, lose, read, write, ...
– p. 16/2
System call example
• Standard library implemented in terms of syscalls
- printf – in libc, has same privileges as application
- calls write – in kernel, which can send bits out serial port
– p. 17/2
Different system contexts
• A system can typically be in one of several contexts
• User-level – running an application
• Kernel “top half” (called “bottom half” in Linux)
- Running kernel code on behalf of a particular process
- E.g., performing system call
- Also exception (mem. fault, numeric exception, etc.)
- Or executing a kernel-only process (e.g., network file server)
• Kernel code not associated w. a process
- Timer interrupt (hardclock)
- Device interrupt
- Software interrupt
• Context switch code – changing address spaces
– p. 18/2
Transitions between contexts
• User → top half: syscall, page fault
• User/top half → device/timer interrupt: hardware
• Top half → user/context switch: return
• Top half → context switch: sleep
• Context switch → user/top half
– p. 19/2
CPU preemption
• Protection mechanism to prevent monopolizing CPU
• E.g., kernel programs timer to interrupt every 10 ms
- Must be in supervisor mode to write appropriate I/O registers
- User code cannot re-program interval timer
• Kernel sets interrupt to vector back to kernel
- Regains control whenever interval timer fires
- Gives CPU to another process if someone else needs it
- Note: must be in supervisor mode to set interrupt entry points
- No way for user code to hijack interrupt handler
• Result: Cannot monopolize CPU with infinite loop
- At worst get 1/N of CPU with N CPU-hungry processes
– p. 20/2
Protection is not security
• How can you monopolize CPU?
– p. 21/2
Protection is not security
• How can you monopolize CPU?
• Use multiple processes
• Until recently, could wedge many OSes with
int main() { while(1) fork(); }
- Keeps creating more processes until system out of proc. slots
• Other techniques: use all memory (chill program)
• Typically solved with technical/social combination
- Technical solution: Limit processes per user
- Social: Reboot and yell at annoying users
- Social: Pass laws (often debatable whether a good idea)
– p. 21/2
Address translation
• Protect mem. of one program from actions of another
• Definitions
- Address space: all memory locations a program can name
- Virtual address: addresses in process’ address space
- Physical address: address of real memory
- Translation: map virtual to physical addresses
• Translation done on every load and store
- Modern CPUs do this in hardware for speed
• Idea: If you can’t name it, you can’t touch it
- Ensure one process’s translations don’t include any other
process’s memory
– p. 22/2
More memory protection
• CPU allows kernel-only virtual addresses
- Kernel typically part of all address spaces,
e.g., to handle system call in same address space
- But must ensure apps can’t touch kernel memory
• CPU allows disabled virtual addresses
- Catch and halt buggy program with wild accesses
- Make virtual memory seem bigger than physical
(e.g., bring a page in from disk only when accessed)
• CPU enforced of read-only virtual addresses useful
- E.g., allows sharing of code pages between processes
- Plus many other optimizations
• CPU enforced execute disable of VAs
- Makes certain code injection attacks harder
– p. 23/2
Resource allocation & performance
• Multitasking permits higher resource utilization
• Simple example:
- Process downloading large file mostly waits for network
- You play a game while downloading the file
- Higher CPU utilization than if just downloading
• Complexity arises with cost of switching
• Example: Say disk 1,000 times slower than memory
- 100 MB memory in machine
- 2 Processes want to run, each use 100 MB
- Can switch processes by swapping them out to disk
- Faster to run one at a time than keep context switching
– p. 24/2
Useful properties to exploit
• Skew
- 80% of time taken by 20% of code
- 10% of memory absorbs 90% of references
- basis behind cache: place 10% in fast memory, 90% in slow,
usually looks like one big fast memory
• Past predicts future (a.k.a. temporal locality)
- What’s the best cache entry to replace?
- If past = future, then least-recently-used entry
• Note conflict between fairness & throughput
- Higher throughput (fewer cache misses, etc.) to keep running
same process
- But fairness says should periodically preempt CPU and give it
to next process
– p. 25/2